Google's spam policy now explicitly covers attempting to manipulate its AI answers. That does not mean optimising for AI Overviews and AI Mode is risky. It means the manipulative tactics were always spam and are now named as such. This is a practical, white-hat method for growing AI visibility that stays clearly on the right side of the line, with a category-by-category test you can apply to any tactic.
The whole method rests on one distinction: optimisation earns a place in the answer by being the best source; manipulation fakes it. Everything below is a way to stay on the earning side.
Step 1: Apply the two-question spam test to every tactic
Before adopting any AI-visibility tactic, ask two questions. If it fails either, it is spam:
- Would it help a real person who landed on this page?
- Would you be comfortable explaining it to Google in plain terms?
Honest optimisation passes both easily. Manipulation fails at least one. Usually the first, because the tactic exists only to trick the engine. This mirrors Google's own framing: spam is about deceiving users or manipulating Search systems, AI answers included. Use the test as a gate on everything that follows.
Step 2: Map your tactics against the named spam categories
Google names specific spam categories, and a handful map directly onto how people chase AI visibility. Check your programme against them:
| If you are tempted to… | It matches this spam category | Do this instead |
|---|---|---|
| Mass-produce thin pages to blanket a topic | Scaled content abuse | Publish fewer, deeper, genuinely useful pages |
| Publish a promotional list on a rented high-authority domain | Site reputation abuse | Earn inclusion in independent lists on merit |
| Stuff AI-target phrases and fake Q&A into a page | Keyword stuffing | Answer real questions clearly; use honest FAQs |
| Build links purely to force prominence | Link spam | Earn mentions by being worth citing |
| Show crawlers content users never see | Cloaking | Serve one honest version to everyone |
| Add fake reviews or unearned schema | Scaled/thin content abuse | Mark up only content you genuinely have |
Source: category names from Google's Search spam-policy documentation, mid-2026. The pattern is consistent. Every "do this instead" is simply the honest version of the same goal.
Step 3: Handle AI-written content by purpose, not by tool
The most common worry in 2026 is whether AI-assisted content is itself a risk. It is not, on its own. Google's category is scaled content abuse: generating many pages primarily to manipulate rankings, and it applies regardless of who or what wrote them.
Google does not penalise content for being AI-assisted. It penalises content made at scale to game the system rather than help a reader. The tool is irrelevant; the purpose and quality are everything.
So AI-drafted content that is accurate, specific, reviewed by a human, and genuinely useful is fine. What crosses the line is publishing volume for volume's sake. We go deeper on this in does AI-generated content hurt AI visibility. The short version is that quality and intent decide it, not the drafting method.
Step 4: Get into best-lists the earned way
Independent best-of lists are among the most-cited page types in AI answers, so the temptation to game them is real, and it is exactly where site reputation abuse lives (renting a trusted domain to place a promotional roundup). The clean alternative is earned placement: make your product genuinely list-worthy, then get reviewers and publishers to include you on the merits.
- Build a real case. Specific capabilities, evidence, differentiation, that a fair reviewer would rank.
- Reach out to legitimate independent publishers rather than paying for a slot on someone else's authority.
- Keep the coverage fresh, because recently-updated lists are cited far more (see the freshness cliff).
The full playbook is in getting into AI-cited best-lists, and it is deliberately all earned-placement. No rented authority.
Step 5: Invest in the lever no update can penalise
The safest and strongest lever is entity strength built through corroboration: publish genuinely useful, specific work, and earn independent sources that agree with it. Because it works by earning trust rather than faking it, no spam update touches it. It is precisely what the policies reward.
- Be corroborated, not just loud. One self-serving claim is fragile; the same fact echoed across reviews, press, and communities is durable. Including in places like Reddit.
- Be consistent across the web. Matching names, facts, and positioning strengthen your entity; contradictions read as untrustworthy.
- Structure honestly. Clean semantic HTML, real structured data for content you actually have, and answer-first writing help engines lift your facts. Legitimately.
Step 6: Measure so you can prove it is working honestly
White-hat visibility compounds slowly, so measure it as a trend rather than expecting an overnight jump. Track your AI citations and AI-referred traffic across engines over several weeks, and watch for the durable signal. Being cited and recommended more often as your corroboration grows.
This also gives you an early warning: if a tactic ever produced a sudden, fragile spike, a smoothed trend line will show it collapsing. The signature of something the engines discounted. Watching what every AI engine says about your brand over time, so you can grow visibility on the durable, policy-safe side of the line, is exactly what Buffy Intel is built to measure.